Dynamic and Semantic 3DGS Active Training on Neu3D average across five dynamic scenes
0.9239SSIMFisher Information-driven NBV selection
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Fisher Information-driven NBV selectionselection strategy=Fisher Information (Semantic-aware), lambda=10^-62025.12 | 0.9239 | 28.6191 | 0.1553 | 89.63 | 76.04 | |
| Geom+Semcomponents=Geometric + Semantic information2025.12 | 0.92 | 28.2788 | 0.1596 | 88.85 | 75.71 | |
| Geo+Defcomponents=Geometric + Deformation information2025.12 | 0.9198 | 28.1501 | 0.1578 | 89.05 | 73.67 | |
| FisherRFselection strategy=Fisher Information (Geometric)2025.12 | 0.9186 | 27.6743 | 0.1605 | 88.63 | 75.47 | |
| Fisher Information-driven NBV selection (lambda=10^-5)lambda=10^-52025.12 | 0.9186 | 28.5295 | 0.1584 | 88.98 | 73.74 | |
| Geomcomponents=Geometric information only2025.12 | 0.9186 | 27.6743 | 0.1605 | 88.63 | 75.47 | |
| Magic Momentsselection strategy=semantic covariance2025.12 | 0.9184 | 28.3795 | 0.1581 | 88.9 | 73.5 | |
| Fisher Information-driven NBV selection (lambda=10^-7)lambda=10^-72025.12 | 0.9184 | 28.0638 | 0.1594 | 88.81 | 73.87 | |
| Fisher Information-driven NBV selection (w/o reg.)regularization=none2025.12 | 0.9173 | 28.5853 | 0.1589 | 89.29 | 74.13 | |
| Randomselection strategy=random2025.12 | 0.8831 | 26.7361 | 0.1933 | 88.48 | 72.84 |